{"id":"https://openalex.org/W7204276154","doi":"https://doi.org/10.48550/arxiv.2608.23830","title":"Mitigating Exploration Bias in RL for Multi-Instruction Following","display_name":"Mitigating Exploration Bias in RL for Multi-Instruction Following","publication_year":2026,"publication_date":"2026-08-24","ids":{"openalex":"https://openalex.org/W7204276154","doi":"https://doi.org/10.48550/arxiv.2608.23830"},"language":null,"primary_location":{"id":"pmh:oai:arXiv.org:2608.23830","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23830","pdf_url":"https://arxiv.org/pdf/2608.23830","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.23830","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148305791","display_name":"Mian Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Mian","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088762304","display_name":"Yueqin Yin","orcid":"https://orcid.org/0000-0002-8952-3590","observed_orcids":["https://orcid.org/0000-0002-8952-3590"]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yin, Yueqin","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109786123","display_name":"Kaiyu He","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"He, Kaiyu","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136466379","display_name":"Peilin Wu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wu, Peilin","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022297444","display_name":"Xinlu Zhang","orcid":"https://orcid.org/0000-0001-7668-091X","observed_orcids":["https://orcid.org/0000-0001-7668-091X"]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhang, Xinlu","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010394308","display_name":"Mingyuan Zhou","orcid":"https://orcid.org/0000-0002-4253-2780","observed_orcids":["https://orcid.org/0000-0002-4253-2780"]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhou, Mingyuan","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5111221510","display_name":"Zhiyu Zoey Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I154570441","display_name":"University of California, Santa Barbara","ror":"https://ror.org/02t274463","country_code":"US","type":"education","lineage":["https://openalex.org/I154570441"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Zhiyu Zoey","raw_affiliation_strings":["UC Santa Barbara"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UC Santa Barbara","institution_ids":["https://openalex.org/I154570441"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I154570441"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.27709999680519104,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.27709999680519104,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.2442999929189682,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.22599999606609344,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/exploration-bias","display_name":"exploration bias","score":0.902999997138977},{"id":"https://openalex.org/keywords/instruction-following","display_name":"instruction following","score":0.753000020980835}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8259000182151794},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.7106000185012817},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.4909000098705292},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.46619999408721924},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4174000024795532},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.385699987411499},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.3847000002861023},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.3564000129699707},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.328000009059906},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2946000099182129},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.289900004863739},{"id":"https://openalex.org/C75917345","wikidata":"https://www.wikidata.org/wiki/Q2725298","display_name":"Sampling bias","level":3,"score":0.2694999873638153},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.26649999618530273}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.23830","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23830","pdf_url":"https://arxiv.org/pdf/2608.23830","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.23830","doi":"https://doi.org/10.48550/arxiv.2608.23830","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.23830","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.23830","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23830","pdf_url":"https://arxiv.org/pdf/2608.23830","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204276154.pdf","grobid_xml":"https://content.openalex.org/works/W7204276154.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"RL":[0,69,82,151,162,194],"has":[1,37],"emerged":[2],"as":[3],"a":[4,41,87,135,144,160,203],"powerful":[5],"paradigm":[6],"for":[7],"enhancing":[8],"the":[9,34,52,72,101,110,126,178,191,200],"instruction":[10,130,209],"following":[11,131,210],"capabilities":[12],"of":[13,92,113,193],"LLMs.":[14],"While":[15],"existing":[16],"training":[17,35,83],"recipes":[18,84],"achieve":[19],"substantial":[20],"gains,":[21],"we":[22,119],"find":[23],"that":[24,165,177],"they":[25],"suffer":[26],"from":[27],"exploration":[28,67,127],"bias":[29,44,128],"towards":[30,76,104],"easy":[31,77,106],"instructions":[32,39,60,93,97,107,169],"when":[33],"data":[36],"multiple":[38],"in":[40,129],"prompt.":[42],"This":[43],"is":[45,61,74],"caused":[46],"by":[47,202],"two":[48,122],"main":[49],"reasons:":[50],"1)":[51,141],"policy":[53,102],"model's":[54],"initial":[55],"ability":[56],"to":[57,64,108,124,138,152,168],"satisfy":[58],"hard":[59,154],"too":[62],"low":[63],"trigger":[65],"successful":[66],"during":[68],"training,":[70],"so":[71],"optimization":[73],"biased":[75],"instructions;":[78,155],"and":[79,132,156,187],"2)":[80,157],"canonical":[81],"typically":[85],"employ":[86],"cumulative":[88],"reward":[89,163],"(the":[90],"number":[91],"fulfilled),":[94],"treating":[95],"all":[96],"equally,":[98],"which":[99],"biases":[100],"model":[103,185],"fulfilling":[105],"obtain":[109],"same":[111],"amount":[112],"reward.":[114],"To":[115],"address":[116],"these":[117],"issues,":[118],"first":[120],"propose":[121],"metrics":[123,180],"measure":[125],"then":[133],"introduce":[134],"two-stage":[136],"framework":[137],"alleviate":[139],"it:":[140],"Behavioral":[142],"Bootstrapping,":[143],"lightweight":[145],"rejection":[146],"sampling":[147],"fine-tuning":[148],"stage":[149],"before":[150],"activate":[153],"Scarcity-Aware":[158],"Rewards,":[159],"new":[161],"function":[164],"assigns":[166],"rewards":[167],"based":[170],"on":[171],"their":[172],"empirical":[173],"scarcity.":[174],"Experiments":[175],"show":[176],"proposed":[179],"are":[181],"highly":[182],"correlated":[183],"with":[184],"performance,":[186],"our":[188,196],"methods":[189],"unleash":[190],"potential":[192],"training:":[195],"best":[197],"models":[198],"outperform":[199],"baselines":[201],"significant":[204],"margin":[205],"across":[206],"three":[207],"verifiable":[208],"benchmarks.":[211],"We":[212],"release":[213],"codes":[214],"at":[215],"https://github.com/mianzhang/MulIF.":[216]},"counts_by_year":[],"updated_date":"2026-09-30T09:27:27.532063","created_date":"2026-08-27T00:00:00"}